Toguyeni et al. (2025) U-NET Deep Learning-based Downscaling to Generate High-resolution Seasonal Forecasts for Small Watersheds: A Case Study of the Nouhao Sub-basin, Burkina Faso

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: International Journal of Environment and Climate Change
• Year: 2025
• Date: 2025-12-03
• Authors: Abdérahim Toguyeni, Ali Doumounia, Moumouni Djibo, Wenceslas Somda, Lucien Damiba, François Zougmoré
• DOI: 10.9734/ijecc/2025/v15i125156

Research Groups

Not explicitly mentioned in the provided text.

Short Summary

This study develops a U-Net Deep Learning framework to downscale coarse 1° (~100 km) seasonal forecasts of precipitation and temperature into high-resolution 0.05° (~5 km) data for Burkina Faso, demonstrating substantial skill improvements (up to sixfold for precipitation and twenty-fold for temperature) compared to raw forecasts.

Objective

• To design and assess a Deep Learning-based downscaling framework using a U-Net Convolutional Neural Network (CNN) to transform coarse 1° (~100 km) seasonal forecasts of precipitation and temperature into high-resolution 0.05° (~5 km) gridded data for Burkina Faso, thereby addressing limitations for local decision-making and disaster preparedness.

Study Configuration

• Spatial Scale: Input forecasts at 1° (~100 km) resolution, downscaled to 0.05° (~5 km) resolution, covering the entire country of Burkina Faso (initially applied to the Nouhao sub-basin).
• Temporal Scale: Seasonal forecasts.

Methodology and Data

• Models used:
  • Deep Learning: U-Net Convolutional Neural Network (CNN).
  • Global forecast models: ECMWF, Météo-France, CMCC (and their ensemble mean).
• Data sources:
  • Raw seasonal forecasts: From ECMWF, Météo-France, and CMCC global models.
  • High-resolution reference datasets (for skill assessment): CHIRPS (precipitation) and CHIRTS (temperature).

Main Results

• Raw seasonal forecasts showed weak performance in representing spatial variability, with biases exceeding 20% for precipitation and more than 5 °C for temperature.
• The Deep Learning-based downscaling process substantially improved forecast skill, with gains of up to sixfold for precipitation and twenty-fold for temperature compared to raw forecasts.
• ECMWF was identified as the best-performing model for precipitation, while Météo-France performed best for temperature, based on a modified Taylor diagram.
• The downscaling framework, initially applied to a sub-basin, was extended to cover the entire country of Burkina Faso, providing high-resolution seasonal precipitation and temperature datasets.

Contributions

• Demonstrates the added value of U-Net Deep Learning-based downscaling for generating high-resolution seasonal precipitation and temperature forecasts in Burkina Faso.
• Provides new insights for integrating Deep Learning approaches into operational drought and flood prediction frameworks for Burkina Faso’s National Agency of Meteorology (ANAM) and the General Directorate of Water Resources (DGRE).
• Contributes to an improved understanding of complex seasonal climate hazards across West Africa, enhancing related areas such as hydrological modeling.

Funding

Not specified in the provided text.

Citation

@article{Toguyeni2025UNET,
  author = {Toguyeni, Abdérahim and Doumounia, Ali and Djibo, Moumouni and Somda, Wenceslas and Damiba, Lucien and Zougmoré, François},
  title = {U-NET Deep Learning-based Downscaling to Generate High-resolution Seasonal Forecasts for Small Watersheds: A Case Study of the Nouhao Sub-basin, Burkina Faso},
  journal = {International Journal of Environment and Climate Change},
  year = {2025},
  doi = {10.9734/ijecc/2025/v15i125156},
  url = {https://doi.org/10.9734/ijecc/2025/v15i125156}
}